[CALL FOR TESTING] Netmap generic mode queue stall fixes

[dfw3xam1n3r]

We are looking for internal approval between participating parties on the last published state for 23.1.6. Overall it looks like we are better off with the patches than without and we likely won't get broader feedback otherwise. If not I expect 23.1.7 to have it in a few weeks.

Based on this comment, I was thinking the patches for netmap issues were going to be a part of the 23.1.7 release. Guess that's not the case.

[franco]

Because they were part of 23.1.6 ;)


Cheers,
Franco

[dfw3xam1n3r]

Ohhh. Geez man, I'm slow. Hmm, well I'm wondering then why I'm still getting these drop issues when I switch ZenArmor out of monitoring-only mode. :\ Oh well, thanks for the help.

[franco]

No worries. 23.1.6 seems to work well enough as a base for future improvements. It may be easier to pinpoint cases such as yours now. We are currently collecting feedback to see if another netmap improvement round is viable...


Cheers,
Franco

[dfw3xam1n3r]

Understood.

Question on ZenArmor config: Do I need to use the Emulated driver for this or can I use Native?

[franco]

This fix is only related to emulated mode. Native wasn't being worked on.


Cheers,
Franco

[Arien]

Hi Franco.

Just a question related to netmap and his improvement.
Is anybody working to match netmap with PPPoE and trying to solve Suricata (netmap) IPS not working at all?

The way you told us "Zenarmor and OPNsense have been working with Klara to bring netmap improvements to FreeBSD" made me think its possible that someone is working on it.

https://forum.opnsense.org/index.php?topic=19740.msg92114#msg92114

Thanks for your time.
Cheers.

[almodovaris]

There is some work still going on about kernels, but I don't know precisely what.

[franco]

The PPPoE/mpd5/Netgraph implementation which we currently use is likely never going to support Netmap for a number of technical reasons/challenges.

We worked on tun(4) previously which would enable the native PPPoE software to benefit from Netmap, but here also technical difficulties in the way tun(4) uses pseudo headers and no link layer headers poses a fundamental issue for Netmap integration since these packets cannot be shared as pure Ethernet packets because they never are. The tun(4) patch approach was abandoned during the project therefore.


Cheers,
Franco